Spotlight: Human Browser — bot-check bypass as an API line item
Anyone who has tried to point an AI agent at a real website knows the wall it hits fast: Cloudflare Turnstile, hCaptcha, reCAPTCHA, DataDome, PerimeterX. Headless Chrome on a datacenter IP gets flagged in seconds no matter how clever the prompt is. Human Browser's answer is to stop trying to fake a human in software and instead make the whole stack look like one: real residential IPs (190+ countries, with a published catalogue of 75), a source-patched Chromium that covers canvas, WebGL, audio and font fingerprints at the engine level rather than with a JS shim, plus bezier mouse movement, human-speed typing and natural scroll timing layered on top.
The interesting decision isn't the anti-detection stack itself, plenty of scraping vendors sell some version of that. It's the framing and the plumbing around it. The site's pitch is blunt: stop buying a $600 Mac Mini or paying $500/mo for a cloud browser subscription, run this on a $5 Linux VPS instead, no display server, no VNC, no desktop. Pricing is metered per task and per agentic action (their numbers: $0.13/task, $0.023/agentic), pulled from a prepaid balance that doesn't expire. And it's built A2A-native: an agent can hit GET /api/plans, POST /buy, get payment auto-confirmed, then poll GET /status/:id, all without a person opening a dashboard. That's a small but real bet on a world where agents provision and pay for their own infrastructure, not just use it.
The MCP skill (npx @virixlabs/humanbrowser) and the open-source browser-human.js script are the other half of the pitch, letting you drop this into whatever agent stack you already run, Claude, Cursor, Gemini, Codex, whatever is issuing the vision-LLM clicks. Worth noting from their own feature list: the self-hosted script path requires you to bring your own residential proxy, so the fully-managed 'zero setup' story applies to the cloud service, not necessarily every tier.
Who this is for: teams building agents that need to actually operate on the open web, filling forms, checking availability, pulling data from sites that already fight scrapers, and who are currently either eating the cost of a $500/mo cloud browser or losing agent runs to CAPTCHA walls. If you're already running Docker, GitHub Actions or any CI/CD pipeline and want to bolt browsing capability onto an existing agent rather than stand up new infrastructure, the setup story here is genuinely light.
Who should skip it: if what you need is a browser for a human to click through manually, or a general-purpose scraping platform with dashboards and job scheduling, this isn't that, it's a stealth browsing layer, not a full scraping product (their own comparison page pitches it as a layer to sit under Apify, not a replacement for it). And it's worth being clear-eyed that a product built specifically to defeat Cloudflare, hCaptcha and PerimeterX is, by design, doing something those services are built to stop, which is a legitimate use for a lot of agent workflows but worth thinking through against the terms of service of whatever sites you're pointing it at.
Where I'd guess this goes next: the A2A payment flow is the tell. If agents start routinely provisioning and paying for their own tools, a browser that already speaks that language natively has a head start over one bolted onto a human billing dashboard later. That's a bet on infrastructure, not a feature, and it's the kind of bet that's easy to dismiss now and obvious in hindsight if agent-to-agent commerce actually happens the way the space keeps predicting it will.
Try Human Browser: humanbrowser.cloud
See the launch: Human Browser on welaunch.sh